Channel jobs · TranscriptFetch docs

A whole channel's or profile's transcripts as one background job: start it, follow its progress, and page through its results.
Beta. A channel job fetches the transcripts of a whole YouTube channel or TikTok or Instagram profile as one background job: the call for loading a creator into a knowledge base or a RAG index. Start it with one request, follow its progress, page through its results, and let a webhook tell you when it ends. It costs what the same work by hand costs: the listing pages and each transcript.
The answer is 202 at once, with the job queued. channel takes anything the channel endpoint takes: a YouTube @handle, UC… id or channel URL, or a TikTok or Instagram profile URL. The platform is read from it.
mode is captions unless you ask for auto, so a job never starts paid AI transcription by surprise. With auto, a captionless video with speech goes to AI Fallback Transcription, and the job waits for it before it completes.
status is queued, running, completed, failed or cancelled. listed counts the videos found so far, completed and failed the transcripts that came back or could not, processing the AI transcriptions still running, and credits_spent what the job has cost.
results holds 25 results at a time in listing order (limit takes up to 50); send next_cursor back as cursor for the next page. Each result is shaped like a batch entry, with the job's transcript options applied:
The words come from the transcript cache, so a job's results stay readable for 7 days after it ends (results_expire_at). Read them into your own store within that time; a result whose words have expired answers expired: true.
With webhook_url, the job POSTs channel_job.started when it begins and channel_job.completed or channel_job.failed when it ends. The body carries job_id, status, the counters, error and results_url, and it is signed like a monitor delivery: webhook_secret from the 202 is the key, and Verify the signature shows how in Node and Python. Delivery is best effort; the job itself is the record, so poll it if a webhook does not come.
A job is billed as it goes, at the API's normal prices: 1 credit for each listing page that has videos, 1 for each caption transcript delivered, and AI Fallback Transcription 1 credit per started 5 minutes of audio in auto mode. Failures are free, and starting, reading and cancelling a job cost nothing. A channel of 500 captioned videos costs 510 credits: 10 pages and 500 transcripts.
Starting a job is refused with 402 insufficient_credits when the account cannot pay for the first page, with 400 invalid_request when max_videos is over the plan's limit, and with 409 job_limit_reached while the plan's number of jobs is already running.
Every call for starting and reading channel jobs: parameters, request and response.
Fetch a channel's or profile's transcripts as one job: it lists the channel a page at a time and fetches each page's transcripts in the background, then calls your webhook. Answers 202 at once; poll the job for progress and read its results a page at a time. Beta.
The account's 20 most recent channel jobs, newest first, without their results. Free.
Progress (status queued, running, completed, failed or cancelled; listed, completed, failed, processing, credits_spent) and a page of results in listing order, each shaped like a batch entry with the job's transcript options applied. Free.
Stops a queued or running job before its next step; nothing more is charged, and results fetched so far stay readable. A finished job is returned unchanged. Free.